提出了一种用多层神经网络对图像进行非线性分割的方法，讨论了所用多层神经网络的学习速度的改进与训练样本的选择方法。实验表明，该多层神经网络系统可用于实时图像分割，并能获得很好的结果 A multi-layer neural network is proposed to segment images non-linearly. The improvement of learning speed and selection of training samples for multi-layer neural networks are discussed. Experiments show that the multi-layer neural network system can be used for real-time image segmentation, and can get good results